Размещено на http://www.allbest.ru/

Размещено на http://www.allbest.ru/

FEDERAL STATE AUTONOMOUS EDUCATIONAL INSTITUTION FOR HIGHER EDUCATION «NATIONAL RESEARCH UNIVERSITY

«HIGHER SCHOOL OF ECONOMICS»

Faculty «Saint Petersburg School of Economic and Management»

Department of Finance

Identification of natural boundaries and communities in St. Petersburg bachelor thesis

EDUCATIONAL PROGRAMME 38.03.01 «Economics»

Admiralov Daniil

Saint-Petersburg

2020



ФЕДЕРАЛЬНОЕ ГОСУДАРСТВЕННОЕ АВТОНОМНОЕ ОБРАЗОВАТЕЛЬНОЕ УЧРЕЖДЕНИЕ ВЫСШЕГО ОБРАЗОВАНИЯ «НАЦИОНАЛЬНЫЙ ИССЛЕДОВАТЕЛЬСКИЙ УНИВЕРСИТЕТ «ВЫСШАЯ ШКОЛА ЭКОНОМИКИ»

Факультет «Санкт-Петербургская школа экономики и менеджмента»

Департамент финансов

Выпускная квалификационная работа

Выделение естественных границ и сообществ Санкт-Петербурга

по направлению подготовки 38.03.01 «Экономика»

Адмиралов Даниил Владимирович

Санкт-Петербург

2020



Introduction

Municipalities in the modern city play a crucial role in citizens' life - landscaping is carried out; municipal deputies listen to the residents of their community and implement initiatives aimed at improving overall community life. Effective municipal management is one of the key tasks in city management. In European countries, municipalities are considered the oldest and most enduring elements of European administration (Rudie Hulst & Andrй van Montfort, 2006).

The city of St. Petersburg includes 111 municipalities. Municipalities include not only urban districts, but also separate settlements, such as the city of Pushkin or the village of Metallostroy. In such settlements, residents often constitute a cohesive community with similar interests. For example, the Saperny settlement consists of several five-floor houses, a couple of dozen private buildings, and the only large enterprise - "Baltica" brewery. However, in the case of St. Petersburg, many municipalities can hardly be called full-fledged communities, since the interests of residents often do not coincide. According to the results of monitoring the development of municipalities of St. Petersburg in 2019, the share of the population that took part in leisure activities (provided by municipality itself) in the municipality of Komendantsky Aerodrom is 49.9%, that is almost half of the total population. At the same time, a similar indicator in the municipality of Shuvalovo-Ozerki amounted to only 1.5%. What is the nature of this difference? If you look at the municipalities' territory Municipalities' territory on the map can be seen in Appendix 1, you will notice that the Komendantsky Aerodrom includes many apartment buildings around the Pionerskaya metro station and the territory of the city park Udelny. Shuvalovo-Ozerki includes not only apartment buildings, but also 3 large lakes on the other side of the Vyborg highway, with many private houses along the perimeter. The interests of the "multi-apartment" part of the district may not coincide with the interests of private-sector residents.

Some municipalities are separated by railways or even rivers, some at the same time include both private and apartment buildings. One of the indicators of community connectivity is the turnout at the municipal elections - the higher it is, the higher is the interest of people in the leadership of the municipality. In St. Petersburg, the turnout for municipal elections between individual municipalities is often significantly different. So, in the elections of municipal deputies of the 4th convocation in 2009, the turnout in the municipality of Tдrlevo Village to more than 96%, and in “Municipal District No. 57” is determined to a little more than 65%.

The purpose of this study is to form a map of possible municipalities of St. Petersburg based on natural boundaries.

As part of the study, the following tasks will be performed:

1. Formation of a new map of the municipalities of St. Petersburg based on natural boundaries using different cluster analysis algorithms

2. Comparison of the created map with existing municipal boundaries

3. Analysis of the results

Chapter 1. Municipalities: from occurrence to optimization



The emergence of municipalities

The main research question of this paper: "How natural are the existing boundaries of municipalities?" To answer this question, it is necessary to turn to the history of the emergence of local communities in Russia. In the 19th century, the entire territory of the empire was divided into provinces and counties, which, in turn, were divided into church parishes. “An Orthodox parish refers to a meeting of Orthodox Christians, who are part of the flock of a local bishop who lives in a certain area near his church and is handed to the nearest pastoral leadership of one or several priests to achieve eternal salvation through common prayer, grace sacraments, edification, and works of Christian charity” (Ivanov P.A., 1914). Life in such parish presupposed the existence of rights and obligations, and within its borders, shepherds and flocks respected their interests. One of the features of the St. Petersburg province was the heterogeneity of counties in terms of social and economic development. Thus, Petersburg and several other counties significantly outstripped the peripheral counties in the south-west of the province. Nevertheless, such communities assumed territorial proximity to the church, which ensured the neighborhood of all its inhabitants. The residents of the parish were essentially a cohesive community of people since the neighborhood and similar life often formed common interests among all residents.

The city of St. Petersburg itself was divided into police units, on the basis of which the administrative division of the city was carried out. In 1917 they were liquidated, and on their basis administrative districts of the city were formed, which were subsequently fragmented into modern municipalities.

The modern system of local government in Russia started to take shape at the turn of the 1990s. The process of the formation of this system was in an atmosphere of the growing crisis, and then the collapse of Soviet statehood. Russia was in a deep state crisis connected with the weakening, and then the collapse of one of its most important components - the “vertical” of the CPSU party power. At the same time, administrative structures of state power in the localities were mainly preserved. Nevertheless, the entire political system of the country was in this period in a state of dynamic change and uncertainty (Avdonin, V., 2008)

The formation of local self-government took place at the beginning by expanding the competence of local councils, defining the scope of their independent tasks, and introducing their managerial autonomy. At the same time, the prerequisites for democratic, free and competitive elections to local authorities were being created. This trend was opposed by the Soviet tradition of centralization, aimed at maintaining administrative control over local councils and subordinating them to higher executive bodies.

In 1993, supporters of strengthening the federal center associated with the president prevailed in Russia; as a result, local self-government bodies disappeared and recovered only in 1997, and in 1998 in St. Petersburg. In the same year, state authorities (the governor and the city administration) won in St. Petersburg. As a result, a micro-district (“quarterly”) model appeared. This method of forming municipalities certainly did not imply ensuring the neighborhood of all residents of the community or provide coherence of interests within the intended community.

This markup of the city led to the extremely low interest of city residents in municipal elections. In 2004, the turnout threshold was introduced in the municipal elections of deputies of the 3rd convocation. The elections had to be held again, as in most municipalities they did not take place due to the established threshold.

Municipalities comparison

The development of city municipalities is assessed annually by the Council of municipalities of St. Petersburg with the support of St. Petersburg administration. The implemented assessment system is supposed to estimate the social and economic development of municipalities along with the effectiveness of local authorities. Based on 2017 monitoring results a comparative analysis of three municipalities was performed:

Indicator	Svetlanovskoye	Chkalovskoye	Kronstadt
Area, sq. km	22,03	9,50	19,35
Population	85243	28729	44461
Income per resident, rub	1538,64	5735,84	2849,89
Residents took part in leisure events, %	10,5	21,1	14,0
Residents took part in sports events, %	0,4	7,8	5,3
Residents took part in celebration events, %	20,2	3,7	8,1
Residents took part in patriotic events, %	0,1	0,8	20,5
Total cost of events per resident, rub	126	210	84
Budget share for the maintenance of local governments, %	20,7	62,8	22,4
Turnout on municipal elections in 2014, %	71,8	64,2	78,7

The largest turnout in the municipal elections was shown by Kronstadt, an isolated city All three municipalities' territory can be seen on the map in Appendix 2, most of which do not contain residential buildings. The historical past of Kronstadt explains the highest rate of residents that took part in patriotic events and may partly reflect the interests of local residents.

The name of Svetlanovskoye municipal district, as well as the names Svetlanovskaya square and Svetlanovskiy Prospekt, are associated with the name of one of the largest enterprises in the Vyborgsky district - the Svetlana plant. This and other industrial enterprises represent the main job providers in this municipality. Thus, we can assume the possible presence of a coherent community in this municipality - employees of industrial enterprises, which is confirmed by a relatively high turnout rate.

The lowest rate of turnout among the municipalities was shown in the municipal district of Chkalovskoye. This can be explained by the geographical location of the municipality - the district consists of several isolated islands connected by bridges. A large number of natural borders within the municipality can lead to a loss of connectivity of the local community and cause a low level of interest in the elections of municipal deputies. A large number of barriers can cause difficulties in the management of the municipality and consequently induce the low efficiency of such management. One of the possible signals of ineffective management may be an indicator of the share of the budget spent on maintenance of the local government itself. In the case of Chkalovskoe, the majority of the municipal budget (62.8%) in 2017 was consumed by the maintenance, which may indicate the need to change the management system or divide the district into several connected communities.

municipality cluster analysis

Related Literature

One of the earliest studies of municipalities (Heikkila, E.J., 1996), examines Los Angeles districts aiming to reveal empirical evidence of the cohesion of local communities. The research showed that existing municipalities often include local communities reflecting a specific population group. The author suggests that municipalities may offer convenient policy levers, such as zoning and other land-use control measures, that can enhance the hallmarks of residents of individual municipalities. However, results show that municipal boundaries change much more slowly than socio-demographic factors. The map of Los Angeles County and its constituent municipalities remained virtually unchanged from what it was a couple of decades ago. The author concludes, that such an observation reinforces the analogy with shopping centers, where the external walls and internal partitions are relatively durable, and the rental turnover from year to year reflects changes in demography.

The study of the effectiveness of municipalities has become a popular theme for a large number of scientific papers. Exploring municipal expenses in Portuguese communities Antуnio Afonso and Ana Venвncio (2016) use data envelopment analysis (DEA) efficiency scores to find out if clustering municipalities into local region clusters would increase the efficiency of municipalities' funds' spending. Authors use hierarchical clustering to create new geographical aggregation. They focus on reducing local government spending, which is a well-known issue in regional economics. The concept of the study is based on inter-municipal flows of working residents. Researchers are considering Portugal as their study area showing incredible results of possible approach implementation: 83 to 98 percent of local government could improve their spending efficiency after the following proposed aggregation. The outcome of the research is robust to different cluster criteria.

This paper was preceded by another research by Antonio Afonso - this time with Sonia Fernandes. They applied DEA methodology to Portuguese municipalities to assess their ability to improve output without spending increase. By comparing the averages of input efficiency scores authors concluded that the Alentejo region and Algarve could theoretically achieve on average roughly the same level of local output with about 34.6 and 39.2% fewer resources, respectively. By contrast, municipalities belonging to the Centro region are reported as being on average the least efficient, implying that these municipalities could theoretically achieve on average roughly the same level of local output with about 76.3% fewer resources, which means that local performance could be strongly improved without necessarily increasing municipal spending. Another significant article was performed by AssunЗгo, R. M., et al. in 2006. Instead of aggregating municipalities to local clusters, authors suggest regionalization based on spatial objects' division. They propose using a connectivity graph approach along with a minimum spanning tree (MST), which is a connected tree with no circuits. MST is being partitioned by removing edges that connect dissimilar communities. The presented approach is considered as an efficient way for optimized regionalization and is called SKATER (Spatial `K'luster Analysis by Tree Edge Removal). Another essential aspect of the study is that the proposed algorithm allows adding restrictions to the procedure. The algorithm is also available as an open-source solution that allows using the approach by anyone.

Although spatial distribution of communities is frequently considered as the main aspect of the municipality's efficiency, other substantial factors influence the local community, such as environmental conditions or level of local government power. All these criteria are considered in the study of Spanish municipalities presented by Marнa Teresa Balaguer-Coll, et al. in 2012. Authors modified the approach of Cazals et al. (2002) to control for the existence of municipalities facing different conditions. The improved method allowed researchers to compare the efficiency of municipalities because of the significant alleviating dimensionality problem. The results of the study show that the difference in circumstances has to be considered in the process of measuring effectiveness. The authors conclude that environmental conditions are especially significant for municipalities' effectiveness assessment. The results are also robust to outliers.

Another significant paper estimates the relationship between urban spatial expansion and its socio-economic determinants (Guastella G. et al., 2017). Authors focus on the most urbanized region of Italy - Lombardy. The estimation of municipality performance is based on an econometric framework that is supposed to analyze determinants of the urban spatial expansion accounting for spatial relationships between neighboring municipalities and structural heterogeneity related to their size, as measured by the total population. The results show that the response of communities to population growth increases as the size of the municipality decrease. The authors also found evidence of spatial effect in the geographical distribution of urbanization. One of the most important results' implications is the high possibility of inefficient land-use planning followed by excessive administrative fragmentation. The authors suggest improving the coordination of the land planning policies between municipalities' administration to hold up the land transformation and soil sealing in general.

Speaking in terms of Russia, in the study by Tatiana Kuznetsova in 2015, the analysis of geo-demographic development of Kaliningrad region is executed using cluster analysis. She classified municipalities according to economic and socio-demographic criteria along with population distribution figure using hierarchical and k-means clustering analysis. The results of the study revealed that the main engine of regional development is the regional center which is simply pulling out resources from other municipalities. Tatiana Kuznetsova also warns about the necessity of appropriate regional policy which should be directed towards avoiding increasing divergence process. The significance of this warning is provided by the high possibility of migrating human resources to the regional center from the periphery.

Chapter 2: Searching for natural boundaries. Research Methodology

During the study, St. Petersburg will be divided into connected municipalities using clustering algorithms. In terms of church parishes, all its residents were supposed to live in a close neighborhood where all the neighboring houses were within walking distance. Hence, in this study it was decided to consider the time of the pedestrian way as metric of connectivity.

The study includes several stages:

1. Data collection and preparation

2. Drawing a grid on a map of St. Petersburg

3. Identification of the boundaries of possible municipalities based on natural boundaries using clustering algorithms

4. Creating a map of possible municipalities of St. Petersburg based on the data obtained and comparing the results of the algorithms

5. Comparison of the created map with existing municipalities

The natural boundaries for a pedestrian in an urban environment are not only rivers or terrain but also man-made obstacles, such as railways or highways. In this study, the final time of the pedestrian path is used as a metric of proximity, taking into account all possible barriers.

As part of the work, the following tools are involved:

- OpenStreetMap (OSM) - a source of open map data used to build walking routes between points;

- Docker-compose - application container used to create a local OSM server

- NextGIS is a source of spatial data; it is used in the work as the main source of spatial data for St. Petersburg;

- Python is a programming language used at all stages of work.

Data Structure

At the first stage of the work, it was necessary to get St. Petersburg data in the form of.shp files containing spatial layers of the city, for example, a layer of land borders or a layer of buildings. The layers were obtained from NextGIS under the ODBL license. The received information contained descriptive data about each individual layer object with an indication of the geographical coordinates of the object. Thus, data contained several variables:

NAME - object name

AMENITY, BUILDING - a type of object, for example 'school', 'bank' or 'road'

GEOMETRY - geospatial information about an object, polygon or point on the map

FEATURES - additional parameters of the object (for example, the type of road), the set of options depends on the layer.

The data type is spatial data. The unit of observation is one spatial object.

Grid Mapping

As part of the tasks, it was necessary to divide St. Petersburg into separate blocks, each of which was part of a common grid superimposed on a map of the city. It was decided to perform the partition into hexes - regular hexagons. The diameter of one hex was 450 meters. The diameter length was chosen based on technical capabilities: 450m was the optimal diameter length allowing server computing in a reasonable amount of time. Using the developed function of hexes applying, a map of St. Petersburg was formed, divided into 6322 individual elements. Each element was either a hexagon or part of it, in case of the water surface partial covering or laying outside of city borders.

Picture 1. Grid map of St. Petersburg

On the basis of the created grid map, a map of possible municipalities will be formed. The coordinates of each vertex inside each hex of the grid were obtained using Python. Based on received coordinate sets, the coordinates of the centroids of each hex were obtained.

The centroid coordinates were calculated using the following formula:

- coordinates of the centroid of a single hex;

i - hex number;

j - hex vertex number;

, - vertex coordinates;

N - total number of hex vertices.

The created set of centroids will be used in the next step of the research. Next, the time of the pedestrian path between all centroids will be found using the deployed OpenStreetMap local server. The resulting matrix of pedestrian distances will be used furtherly in the clustering process.

Searching for walking routes

For clustering based on proximity, it is necessary to obtain a matrix of distances between the centers of all elements. Many algorithms can cluster spatial data based on their coordinates using Euclidean distance between points. However, the Euclidean distance is not representative in terms of pedestrian accessibility, since neighboring points on the map can be separated by a river or railway, which greatly complicates the movement between these points. In the framework of this study, it was decided to use another metric - the walking time between the points. Thus, to use clustering algorithms, it is necessary to obtain a pedestrian distance matrix.

To compile such a matrix, it was decided to deploy a local OpenStreetMap server containing St. Petersburg's road paving graph with all footpaths. The server was created using the Docker-compose application container, and was available through OSM api, allowing to calculate the pedestrian distance. The OSRM package for the Python language was used to form the matrix. Using this package, the time of the shortest pedestrian path between all points was calculated. The received matrix was used next up in the clustering process.

Clustering and creating a new map of possible municipalities

As part of the study, two algorithms were used to cluster spatial data:

- DBSCAN - Density-Based Spatial Clustering of Applications with Noise

- Agglomerative Clustering

The DBSCAN algorithm considers clusters as high-density regions separated by low-density regions. Due to the generalized representation, the clusters found by DBSCAN can be of any shape, unlike the k-means algorithm, which assumes that the clusters are convex. The central component of DBSCAN is the concept of core elements, which are elements located in high-density areas. Thus, a cluster is a set of elements, each of which is close to each other (measured using a proximity measure), and a set of non-core elements that are close to the core but are not the core ones. The algorithm has two parameters, min_samples and eps, which formally define the concept of “density”. A larger minimum set of elements in a cluster (min_samples) or a lower minimum distance (eps) indicate a higher density needed to form a cluster.

The Agglomerative Clustering algorithm performs hierarchical clustering using a bottom-up approach: each observation begins in its own cluster, and the clusters are combined sequentially. Linkage criteria determine the indicator used for the approach to clustering elements:

- Ward minimizes the sum of squared differences across all clusters. This is an approach that minimizes variance which is similar to the k-means objective function but is solved using the agglomerative hierarchical approach.

- Complete linkage minimizes the maximum distance between observations of pairs of clusters.

- Average linkage minimizes the average distance between all observations of pairs of clusters.

- Single linkage minimizes the distance between the closest observations of pairs of clusters.

During the research all algorithms were implemented using Python. The result of each algorithm was an array, each element of which reflected the belonging of an individual hex to the resulting cluster (possible municipality).

The DBSCAN algorithm has generated extremely long clusters, which can be explained by the eps parameter, which reflects the maximum distance between two samples for one to be considered as in the neighborhood of the other. In the clustering process using this algorithm, two points are recognized as neighboring if it is possible to lay a route between them through other points where the maximum distance between two points does not exceed eps. The most effective was agglomerative clustering, which was able to create a map of St. Petersburg, dividing the city into municipalities, taking into account the proximity of the pedestrian path. The given linkage to the algorithm was `Average', which means that agglomerative clustering determines to minimize the average distance between clusters.

Analysis of the created map

The finally created map is based on natural boundaries and represents the work of the agglomerative clustering algorithm. The map is split into colored blocks that represent separated municipalities with the minimal pedestrian path within the block. During the work, the agglomeration clustering algorithm divided St. Petersburg into a different number of municipalities, from 50 to 120. The algorithm showed the most connected picture in the case of 110 municipalities. In the case of a smaller number the algorithm formed larger clusters that were difficult to access through the pedestrian crossing. In the case of a larger number, the algorithm over fragmented the city, creating tiny difficult-to-interpret clusters.

It was decided to compare the received map of municipalities with the map of the existing administrative division:

Picture 2. Existing municipalities over algorithmically created ones

Results showed that some of the existing municipalities are repeating the forms of algorithmically created ones. This could probably happen because many modern districts, especially in the suburbs, are actually based on the boundaries of rivers. Thus, we can conclude that the boundaries of some administrative units are natural, although this does not mean that they are optimal. If you look at Kronstadt, you can see that the algorithm divided it into two large municipalities. This is no coincidence - exactly in the middle of the city passes the Ring Road, dividing the municipality into two parts. One may notice some errors in the partitioning, which probably indicates the necessity of dividing St. Petersburg into smaller hexes, which will significantly increase the accuracy of clustering, while greatly increasing the computation time.

Conclusion and Discussion

Speaking in terms of Russia, the administrative-territorial division of urban space is often illogical or does not take into account the features of the landscape of the area. In this study three spatial clustering algorithms were executed to form the map of St. Petersburg possible municipalities on the basis of the footpath time. The results show that agglomerative clustering seems to be the most appropriate algorithm among considered for this kind of research. The colored blocks on the map represent algorithmically created possible municipalities based on natural boundaries. The created map was compared with the map of existing municipalities. The comparison revealed, that some of the modern municipalities' boundaries repeat the ones received from the clustering process. However, there were some mistakes clearly visible on the colored map. The outliers were probably appeared due to the large diameter of the hexagon, which will be adjusted in the future.

However, the map created does not take into account the socio-economic, cultural, and other vital factors that affect community connectivity. Thus, the key priority of further research is to evaluate the effectiveness of the formed municipalities on the basis of various economic indicators, as well as an empirical study of sociocultural factors that affect the interests of the community. In the framework of the study, pedestrian path time was used as a primary metric. Despite the fact that clustering on this indicator can provide pedestrian accessibility of facilities within the municipality, this approach does not take into account many factors that shape the interests of community residents. Further studies of this issue suggest a modification of the map, taking into account the necessity to include a mandatory set of socio-economic benefits in each municipality. To improve the administrative-territorial division of the city, it is necessary to conduct a field analysis of local communities, as well as development density analysis.



References

4. Afonso, Antonio and Fernandes, Sonia, (2008), Assessing and explaining the relative efficiency of local government.

5. Afonso, Antonio and Venвncio, Ana, (2013), The relevance of commuting zones for regional spending efficiency.

6. Balaguer-Coll, M., Prior, D., & Tortosa-Ausina, E. (2013). Output complexity, environmental conditions, and the efficiency of municipalities. Journal of Productivity Analysis, 39(3), 303-324.

7. Guastella, G., & Sckokai, P. (2017). (Rep.). Fondazione Eni Enrico Mattei (FEEM). A Spatial Econometric Analysis of Land Use Efficiency in Large and Small Municipalities

8. Heikkila, E.J., (1996), Are municipalities Tieboutian clubs?

9. Heinz, W. (2000). Interkommunale Kooperation in Stadtregionen: das Beispiel der Bundesrepublik, Deutschland, in: Werner Heinz (Ed.), Stadt & Region - Kooperation oder Koordination, Berlin,p. 169-274.

10. Kuznetsova, T.Y. (2015). Geo-Demographic Typology of Municipalities of the Kaliningrad Region.

11. Martins, Assuncao & Neves, Marcos & Cвmara, Gilberto & Da Costa Freitas, Domingos. (2006). Efficient Regionalization Techniques for Socio-Economic Geographical Units Using Minimum Spanning Trees. International Journal of Geographical Information Science. 20.

12. Rudie Hulst, Andrй van Montfort. (2006), Inter-Municipal Cooperation in Europe

13. Schubert, Erich; Hess, Sibylle; Morik, Katharina (2018). The Relationship of DBSCAN to Matrix Factorization and Spectral Clustering, 330-334.

14. Абрамов, В.Ф., (2011), Демократическая практика россии?ского земства

15. Авдонин, В. С. (2008). Развитие местного самоуправления в Германии и России: история, проблемы, перспективы. Политическая наука, (3), 88-110.

16. Иванов П. А., (1914), Историко­каноническое исследование о православном русском приходе в связи с предложенной реорганизациеи? его на древнерусских началах.

17. Комитет территориального развития Санкт-Петербурга, (2015-19), Результаты мониторинга социального и экономического развития внутригородских муниципальных образовании? Санкт-Петербурга и оценки эффективности деятельности органов местного самоуправления внутригородских муниципальных образовании? Санкт-Петербурга по итогам 2015-19 года.



Appendix 1: Komendantsky Aerodrom and Shuvalovo-Ozerki territory

Picture 3. Shuvalovo-Ozerki territory

Picture 4. Komendandsky Aerodrom territory

Appendix 2: Svetlanovskoe, Chkalovskoe and Kronstadt territory

Picture 5. Kronstadt territory

Picture 6. Svetlanovskoe territory



Picture 7. Chkalovskoe territory



Appendix 3: Clustering results

DBSCAN

Picture 8. DBSCAN best results

Agglomerative Clustering

Picture 9. n_clusters=50



Picture 10. n_clusters=60

Picture 11. n_clusters=70

Picture 12. n_clusters=80

Picture 13. n_clusters=90

Picture 14. n_clusters=100

Picture 15. n_clusters=110

Picture 16. n_clusters=120

Размещено на Allbest.ru